saudnr33 / tpti

TPTI: Temporal Pivotal Tuning for Latent-based editing of Real Images.

This repository is built on,

1. EG3D.
2. PTI.
3. Deep3DFaceRecon_pytorch.

Please, follow the intstructions carefully when installing EG3D.

Preparing datasets

Run the following commands:

cd dataset_preprocessing/ffhq
python preprocess_in_the_wild.py --indir=INPUT_IMAGE_FOLDER

pretrained model

The projector needs vgg16 for loss computation, you can download vgg16.pt from https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/metrics/vgg16.pt and save it to eg3d/network.

Projection:

Run the the following for first frame projection:

cd eg3d
python run_projector.py --outdir=projector_out --latent_space_type w  --network=networks/ffhq512-128.pkl --sample_mult=2  --image_path ./projector_test_data/frame001.png --c_path ./projector_test_data/frame001.npy

PTI projector

Notice: before you run the PTI, please run the w to get the latent code of the first frame.

Then run:

cd eg3d/projector/PTI
python run_pti_video.py --in_dir PATH/TO/DIRECTORY --run_name Experiment_name

You can generate a video of the results using:

python saud_gen_video.py --in_dir PATH/TO/DIRECTORY